supplementary material table s01

1

Supplementary Material

Table S01. List of dates of Landsat-5 and Landsat-8 scenes used in the respective ice regimes in both hemispheres. If there is

more than one scene used per day, the date has an ‘a’, ‘b’, and so forth appended. Note that these are not all scenes used but

only those that fit adequately into our definition of ice regimes. Note further that there is overlap between regime ‘melt

conditions’ and other regimes.

Landsat-5, NH Landsat-8, NH Landsat-8, SH

Freeze-up None 20140918, 20150914,

20150915

20140216, 20140223,

20140227a, 20140227b,

20140305, 20140307,

20140312a, 20140312b,

20150312a, 20150312b,

20150321

# cases 0 3 11

Ice edge None 20130427a, 20130427b,

20130527, 20150316,

20150405, 20150419

20131130, 20131228,

20140114, 20140115,

20140216, 20140227,

20141222, 20141224

# cases 0 6 8

Leads / Openings 20030407, 20050408,

20050410, 20050414a,

20050414b, 20050419,

20050423, 20050521,

20060402, 20060404,

20060408, 20060421,

20060426, 20060427,

20060430, 20070407,

20070409, 20070412a,

20070412b, 20070415,

20080321, 20080407,

20080409, 20080423a,

20080423b, 20090320,

20090418, 20090426,

20090524, 20100510,

20100527

20130425, 20130501,

20130529, 20140321,

20140326, 20140330,

20140402, 20140416,

20140428, 20150401,

20150404, 20150406,

20150407, 20150413,

20150414a, 20150414b,

20150414c, 20150416,

20150422, 20150423a,

20150423b, 20150423c,

20150425a, 20150425b,

20150427, 20150428,

20150514, 20150516,

20150524, 20150528

20131105, 20131109,

20131117, 20131118,

20131208, 20131214,

20131221, 20131227,

20140102, 20140203,

20140215, 20140223,

20140227, 20140307,

20140312, 20141024,

20141030, 20141201,

20141220, 20151002,

20151005a, 20151005b,

20151021, 20151023a,

20151023b, 20151123,

20151202, 20151212,

20151213

# cases 31 30 29

Heterogeneous ice 20050424, 20050524a,

20050524b, 20080511,

20090321, 20090403,

20090416, 20090528,

20100510, 20100524a,

20100524b

20130422, 20140503a,

20140503b, 20140517a,

20140517b, 20150419,

20150420, 20150425a,

20150425b, 20150425c,

20150427, 20150516a,

20150516b, 20150519a,

20150519b, 20150522a,

20150522b

20131103, 20131108,

20131111a, 20131111b,

20131111c, 20131208a,

20131208b, 20131210a,

20131210b, 20131212,

20140101, 20140114a,

20140114b, 20140118,

20140129, 20140305,

20141009, 20141028,

20141031, 20141101,

20141203, 20141205,

20141210, 20141224,

20141226, 20150312,

20151001, 20151022,

20151111, 20151120,

20151128, 20151130,

20151212

# cases 11 17 33

2

Table S01, continued.

Landsat-5, NH Landsat-8, NH Landsat-8, SH

High-concentration 20030403a, 20030403b,

20030403c, 20030405a,

20030405b, 20030405c,

20030416a, 20030416b,

20030416c, 20040315,

20040320, 20040409,

20050410, 20050414,

20050419, 20050421a,

20050421b, 20060404,

20060413a, 20060413b,

20060421, 20070407a,

20070407b, 20070411a,

20070411b, 20070411c,

20080411, 20100510

20140405, 20150404,

20150411a, 20150411b,

20150412, 20150416a,

20150416b, 20150418,

20150420a, 20150420b,

20150420c, 20150420d

20131110, 20131208,

20150312, 20151104

# cases 28 12 4

Melt conditions 20030516, 20040518,

20040522, 20050521,

20050524a, 20050524b,

20060528, 20080520,

20090524, 20090528,

20100524a, 20100524b,

20100527

20130527, 20130529,

20140517a, 20140517b,

20150516a, 20150516b,

20150516c, 20150519a,

20150519b, 20150522a,

20150522b, 20150522c,

20150524, 20150525,

20150528L

20131117, 20131118,

20131130, 20131205,

20131208a, 20131208b,

20131208c, 20131208d,

20131209, 20131210a,

20131210b, 20131212,

20131214, 20131221,

20131227, 20131228,

20140101, 20140102,

20140114a, 20140114b,

20140114c, 20140115,

20140118, 20140129,

20140203, 20140215,

20141201, 20141203,

20141205, 20141210,

20141220, 20141222,

20141224a, 20141224b,

20141226, 20151120,

20151123, 20151128,

20151130, 20151202,

20151212a, 20151212b,

20151213, 20151214,

20151224

# cases 13 15 45

3

Supplementary material to Subsection 2.2.3

Table S02. Landsat SIC using pixels only classified as thick ice derived using the actual pair of albedo threshold values

(“Actual value”) and the four variations of them (see text) averaged for 12 Landsat-8 scenes selected for the Northern

Hemisphere (NH) at 25 km and 50 km grid resolution. The number behind the ± denotes one standard deviation (supplements

Table 3 in the main manuscript).

αthinice \ αopenwater -0.03 Actual value +0.03 NH, 25km

-0.1 94.6 ± 4.8 -- 94.6 ± 4.8

Actual value -- 93.2 ± 4.7 --

+0.1 90.8 ± 4.4 -- 90.8 ± 4.4

NH, 50km

-0.1 94.3 ± 5.5 -- 94.3 ± 5.5

Actual value -- 92.9 ± 5.6 --

+0.1 90.7 ± 5.3 -- 90.7 ± 5.3

Table S03. Landsat SIC using only pixels classified as thick ice derived using the actual pair of albedo threshold values

(“Actual value”) and the four variations of them (see text) averaged for 15 Landsat-8 scenes selected for the Southern

Hemisphere (SH) at 25 km and 50 km grid resolution. The number behind the ± denotes one standard deviation (supplements

Table 4 in the main manuscript).

αthinice \ αopenwater -0.03 Actual value +0.03 SH, 25km

-0.1 55.2 ± 25.2 -- 55.2 ± 25.2

Actual value -- 52.6 ± 24.9 --

+0.1 49.9 ± 24.8 -- 49.9 ± 24.8

SH, 50km

-0.1 47.7 ± 31.1 -- 47.7 ± 31.1

Actual value -- 45.6 ± 30.1 --

+0.1 43.4 ± 29.1 -- 43.4 ± 29.1

Table S04. Comparison of statistical parameters listed in Table 8 for the Southern Hemisphere for SICCI-2 and OSI-450

products and Landsat-8 using truncated and non-truncated (near 100 % SIC) PMW SIC data. See caption of Table 5 for

explanation of the parameters given.

LS8, SH

2013-15

SICCI-

12

SICCI-12

non-

truncated

SICCI-

25

SICCI-25

non-

truncated

SICCI-

50

SICCI-50

non-

truncated

OSI-

450

OSI-450

non-

truncated

Diff -5.1 -4.3 -5.9 -5.6 -6.8 -6.5 -5.3 -5.1

DiffSDEV 13.3 13.8 13.5 13.7 16.0 16.2 13.5 13.7

Slope 0.915 0.931 0.969 0.976 1.033 1.040 0.827 0.832

Intercept 2.1 1.6 -3.3 -3.5 -9.6 -9.9 9.5 9.3

R² 0.78 0.77 0.77 0.77 0.72 0.71 0.73 0.73

4

Supplementary material to Subsection 2.2.4

We assume that a Landsat pixel (30 m x 30 m) is completely covered by either ice or water. For a Landsat pixel

covered just half by snow covered / thick sea ice, which exhibits a surface albedo of 0.8 under cold conditions, the resulting

pixel average albedo is 0.5 x 0.06 + 0.5 x 0.8 = 0.43. With that, the pixel is classified as bare / thin ice and counts as a pixel

with 100 % instead of 50 % sea-ice concentration. Likewise, a pixel covered with 80 % snow covered / thick sea ice with a

similar surface albedo results in an average albedo of 0.2 x 0.06 + 0.8 x 0.8 = 0.65. With that, the pixel is classified as snow-

covered / thick ice and as well counts as a pixel with a sea-ice concentration of 100 % instead of 80 %.

In order to assess the severity of this positive bias it is useful to distinguish between sea-ice conditions during summer

and winter, in the pack ice and the marginal ice zone (MIZ), and to take into account different ice types as well as typical floe

sizes. Regardless of ice type, floe size and location winter conditions are likely less prone to cause overly large biases in the

Landsat SIC unless the sea ice drifts into open water regions with comparably high temperatures. Leads and openings in the

pack ice are mostly frozen over. Only open ones, i.e. not covered by new or thin ice, would have an impact. Leads exhibiting

widths below the Landsat pixel size of 30 meters would not be resolved as open water (see next paragraph). For wider open

leads, the actual open water fraction would be smaller than the actual one because the mixed pixels at their borders would be

classified at 100 % sea ice.

Onana et al. (2013) analyzed airborne digital camera visible imagery captured along Operation Icebridge (OIB) flights

in the Arctic in April and in the Antarctic in October at ~0.1 m nadir spatial resolution for the lead and open water fraction

along the flight tracks; their lengths were ~ 7000 km in the Antarctic and ~ 5300 km in the Arctic. They reported an open water

fraction of ~1.2 % for the Antarctic flight and of ~0.2 % for the Arctic flight. The spatial resolution of their input data of 0.1

m allows to reliably detect 1m-wide leads. These are clearly sub-resolution for Landsat. Lead width distributions follow a

power law with lead counts decreasing with increasing lead width (e.g. Tschudi et al., 2002; Marcq and Weiss, 2012).

Therefore, it can be expected that only comparably few leads along these OIB flight tracks exhibit an open water signature

wide enough to also be identified by Landsat as open water at 30 m pixel size; the bulk of the open water in small-scale leads

remains undetected. With that Landsat SIC is biased high simply because of the “coarse” resolution of the Landsat sensor but

the magnitude of this bias (unknown but smaller than the two above-mentioned percentage values) falls within the uncertainty

range of our approach (Subsection 2.2.3; Tables 3 and 4). For leads wider than 30 m, we can attempt to estimate the contribution

of mixed pixels classified as ice, i.e. the pixels at the lead boundaries, on the Landsat SIC at, for instance, 25 km grid resolution.

Into one 25 km x 25 km grid cell 694 444 Landsat pixels fit. Let us assume there is some land so that the actual ice-covered

ocean area comprises 500 000 pixels. An open water fraction (see above) of ~ 1 % corresponds to 5000 pixels. Landsat cannot

resolve all of these because the bulk of the 1 % open water fraction is from sub-pixel size leads. If we assume that Landsat can

identify half of that open water fraction then we are at 2500 pixels. Of these 2500 pixels two third, i.e. ~ 1600 might actually

be pixels at lead boundaries, hence classified as 100 % ice instead of some fraction between 0% and 100%. When related to

the entire 25 km grid cell and assuming that the average ice fraction of the misclassified pixels is 50 %, these 1600 pixels

correspond to an over-estimation of the SIC by ~ 0.15 %. This value falls into the uncertainty range of our approach (Subsection

2.2.3; Tables 3 and 4). We note that these considerations rely on the assumption that the OIB flights provide a representative

estimate of the conditions of Arctic and Antarctic pack ice during winter conditions.

We also estimated the bias for three typical airborne visible images taken in the MIZ in the Greenland Sea, Arctic,

during winter in March 1997 (http://seaice.dk/esop/egrl97/, last access July 2 2021). The images show a sea-ice cover

comprising closely packed but also broken bands of thicker ice floes, pancake ice, brash and grease ice; the approximate sea-

ice concentration is around 70 %. We superpose the images, which are approximately 4 km wide, with a grid representing the

Landsat pixel size of 30 m x 30 m, and visually count the mixed pixels that exhibit a non-zero open water fraction of

approximately more than ~ 5 %. The number of usable pixels of the 12 060 pixels in total ranges between 10 532 and 11 205

5

because some were obscured by parts of the aircraft and other remote sensing equipment in the field of view of the camera

used. We find between 908 and 2057 mixed pixels, corresponding to a fraction between 8.1 % and 19.2 %. If we translate this

situation to a 25 km grid cell, then our approach would result in between 8.1 % and 19.2 % of the Landsat pixels in that grid

cell classified as 100 % ice instead of some fraction between 5 % and 100 %. By assuming that the ice fraction in these pixels

is, on average, 50 %, we can estimate an overall positive bias in the Landsat SIC of between ~ 4 % and ~ 10 % for the ice

conditions encountered.

During summer, the thin ice detected by Onana et al. (2013) along the above-mentioned OIB flights would be open

water; the corresponding open water fraction would be 5.5 % for the Antarctic flight and 1.5 % for the Arctic flight. We can

expect that Landsat would identify only a comparably small part of these open water patches because of the reasons laid out

earlier, leading to Landsat SIC biased high simply due to the resolution. In addition, mixed pixels at the lead boundaries

classified as 100 % ice would lead to an elevated positive bias. We can repeat the earlier exercise and provide an estimate of

the impact of these mixed pixels. For that we replace the values of the open water fraction of ~ 1 % by ~ 5 %. Note that we

take the Antarctic example here because only in the Antarctic we work with summer-time Landsat scenes. For the conditions

chosen, i.e. ~ 5 % open water fraction, the bias due to mixed pixels would then be the five-fold value of the above-mentioned

one: ~ 0.75 %. We now could expand this examination to even larger open water fractions. However, it is not clear whether it

is justified to linearly increase the value of the potential mixed-pixel bias as a function of the open water fraction, i.e. whether

this bias would amount, e.g., 1.5 % for an open water fraction of 10 % and 3 % for an open water fraction of 20 %. We attempt

to shed light on this issue in the following paragraphs.

A sea-ice cover comprises a mélange of individual ice floes that are frozen together in winter but are often floating

individually, only “glued” together by slush and brash ice formed from disintegrating floes or ridges in summer. Like the lead

width distribution does also the floe size distribution follow a power law with maximum floe count for small floes, decreasing

to smaller counts for larger floes (e.g. Steer et al., 2008; Toyota et al., 2011; Perovich and Jones, 2014). Depending on season

and location also the floe size could cause biases in Landsat SIC estimates based on our approach – primarily because of two

reasons. At first, a considerable fraction of the ice floes encountered in the MIZ, particularly in the Antarctic, exhibit sizes

smaller than the Landsat pixel size of 30 m (Lu et al., 2008; Steer et al., 2008; Toyota et al., 2011; 2016). According to our

first paragraph of Subsection 2.2.4 in the main manuscript, one snow-covered ice floe of ~ 6 m size would be enough to

increase the pixel average albedo above the upper open water / ice discrimination threshold of 0.09 (see Tables 3 and 4) and

to classify that pixel as ice. One ice floe of that size corresponds to a sea-ice concentration of ~ 4 % in that pixel. For an ice

floe with melting snow, bare ice or any form of thin ice without a snow cover the floe size needs to be larger to cross that

threshold; i.e. for thin ice with a surface albedo of 0.35 the floe could be 10 m in size, corresponding to a sea-ice concentration

of ~11% in that pixel. In both cases, the classification results in a pixel with 100 % SIC. A 25 km grid cell comprising such

Landsat pixels would exhibit an extremely overestimated Landsat SIC value. Secondly, for floes larger than the Landsat pixel

size, each floe boundary intersecting with a pixel produces a mixed Landsat pixel. Hence, one isolated ice floe of, for instance,

~ 50 m diameter could, in the worst case, fill one Landsat pixel completely and overlap with the eight surrounding Landsat

pixels as well. As a result, nine Landsat pixels would be classified as ice, i.e. 8100 m² while the actual area of the floe, if

assumed to be a circular disc, would be ~ 2000 m²; this is a four-fold overestimation of the actual sea-ice area.

However, as we state at the beginning of the previous paragraph, the sea-ice cover is a mixture of ice floes of

difference sizes and shapes. During winter, any openings between floes are small and/or frozen; floe size distributions and the

two issues mentioned above do not have a notable impact on the Landsat SIC at the scale of the PMW SIC product grid

resolution – particularly not in the Arctic Ocean (Perovich and Jones, 2014). During summer, the type of ice cover in the

openings is more difficult to assess and to take into account in a quantitative way. According to the high-resolution optical

images shown in the publications used to infer the floe size distribution (Steer et al., 2008; Toyota et al., 2011; 2016) and

similar studies (e.g. Paget et al., 2001; Lu et al., 2008; Zhang and Skjetne, 2015), the ice cover often comprises a large spectrum

6

of floes with small floes sitting in the openings between moderately large floes that in turn sit in the openings between large

floes. In addition, slush and/or brash ice often fills the remaining gaps. Therefore, for these cases the actual sea-ice

concentration is close to 100 % and would also be close to 100 % following our classification method despite the substantial

variation in surface albedo due to the mixture of different ice types.

There are, however, cases where both, floe size and size of the openings between the floes are smaller than the Landsat

pixel size and where in addition the openings are not covered by brash ice or slush, examples of which are shown, e.g., in

Toyota et al. (2011). For these cases, certainly too many Landsat pixels would be classified as 100 % ice and the Landsat SIC

would be biased high at the scale of the PMW SIC product grid resolution. Herding of ice floes under the action of wind waves

and swell can generate larger conglomerates of densely packed floes separated by larger, lead-like openings (Toyota et al.,

2016). The process decreases the amount of isolated sub-pixel scale floes and openings and potentially results in a smaller

positive bias. In the following, we attempt to estimate the efficiency of this herding. We assume two 25 km grid cells, A and

B, in the open ocean, i.e. no land, with an actual sea-ice concentration of 50 %. Grid cell A is covered completely by isolated

floes with a size smaller than the Landsat pixel size, separated by open water patches of the same size. This is a very extreme,

unlikely situation at a scale of 25 km squared. With our approach, we would classify every pixel as 100 % ice and the resulting

25 km grid cell Landsat SIC would be 100 %. Grid cell B is covered by ten parallel bands of herding-induced closely-packed

ice floes with near-100 % SIC separated by open water bands (wide leads). A good estimate for the length of these bands is

1000 Landsat pixels. While our approach would do fine for the near-100 % sea ice band and the pure open water pixels in the

wide leads, the mixed pixels at the wide lead boundaries cause a positive bias – as described already above. Ten wide leads

with two borders on each side estimate to 20 x 1000 = 20 000 Landsat pixels with a mixed surface that would be classified as

100 % ice instead the actual ice fraction between 0 % and 100 %. This amounts to ~ 3 % of the pixels in grid cell B, translating

into an over-estimation of the SIC by ~ 1.5 % under the assumption that the average actual ice fraction in the misclassified

pixels is again 50 %. Hence, for grid cell A, Landsat SIC is 100 % instead of 50 % while for grid cell B Landsat SIC is ~ 52

% instead of 50 %. This example illustrates that for our approach there is a large range of Landsat SIC overestimation at PMW

SIC product grid resolutions for those summer ice conditions where gaps between ice floes are not filled by slush, brash ice or

young ice types.

A special ice form not yet treated is pancake ice. This ice type forms predominantly at the ice edge / in the MIZ from

grease and slush ice under the action of wind and waves. Floe sizes of pancake ice are small, mostly below 10 meters. Despite

this, we are confident Landsat SIC match the actual sea-ice concentration in pancake ice regions well because of several

reasons. Such regions occur during winter conditions and therefore any sub-pixel scale open water fractions are very low.

Pancake ice floes are often embedded into a matrix of grease and slush ice, both exhibiting an albedo above the open water –

ice discrimination threshold used. Actual sea-ice area fractions in pancake ice dominated areas are often near 100 % according

to visual ship-based sea ice observations (e.g. Ozsoy-Cicek et al., 2011; Alberello et al., 2019). Only at the border of pancake

ice regions that often occur in bands an over-estimation of the Landsat SIC due to the above-described mixed-pixel effect is

possible. This effect is more difficult to quantify as in case of the leads (see above) but we doubt that the resulting bias would

exceed 2 % at the 25 km scale of the PMW SIC grid – unless the ice is advected into warmer waters inhibiting ice formation;

then biases could be as large as 10 %.

One way to reduce the unknown bias in the used Landsat SIC estimates could be to switch from the binary

classification, i.e. water or ice, to an estimation of the actual SIC in every pixel as a function of the albedo, i.e. a gradient

approach (see e.g. Kern et al., 2003). That way one would have 0 % SIC for pixels with a typical open water albedo, 50 % SIC

for pixels with an albedo of, e.g. 0.4, and 100 % SIC for pixels with an albedo of 0.8. This is, however, an over-simplification

of the situation as well because an albedo of 0.4 might indeed originate from a pixel with 50 % SIC but it might as well

originate from a pixel covered by 100 % thin ice. Hence, such a gradient approach would result in an unknown amount of

under-estimation of the Landsat SIC in some grid cells. One solution to avoid this negative bias could be to use the albedo of

7

a pixel covered by 100 % thin ice, e.g. 0.4 or similar, as the end-member of the gradient approach, i.e. all pixels with an albedo

> 0.4 are set to 100 % SIC while intermediate SIC is estimated for pixels exhibiting smaller albedo values. However, in order

to generate such a high-quality Landsat SIC data set one needs a more accurate correction of the atmospheric influence, and

consideration of sun elevation and observation angles than employed in this study. In addition, also here an independent data

set of an even finer-resolved ice-water distribution is required to assess the quality.

Additional References:

Alberello, A., Onorato, M., Bennetts, L., Vichi, M., Eayrs, C., MacHutchon, K., and Toffoli, A.: Pancake ice floe size

distribution during the winter expansion of the Antarctic marginal ice zone. The Cryosphere, 13(1), 41-48,

https://doi.org/10.5194/tc-13-41-2019, 2019.

Ozsoy-Cicek, B., Kern, S., Ackley, S. F., Xie, H., and Tekeli, A. E.: Intercomparisons of Antarctic sea ice types from

visual ship, RADARSAT-1 SAR, Envisat ASAR, QuikSCAT, and AMSR-E satellite observations in the Bellingshausen Sea,

Deep-Sea Res. II, 58(9-10), 1092-1111, https://doi.org/10.1016/j.dsr2.2010.10.031, 2011.

8

Supplementary material to Section 4

Figure S01. Scatterplots of PMW SIC (y-axis) versus Landsat SIC (x-axis) for SICCI-2 and OSI-450 products for Landsat-8

melt conditions cases during years 2013-2015 in the Southern Hemisphere. Black dots are individual data pairs, the black solid

line is the linear regression, and the black dashed line is the identity line. Red triangles denote the mean PMW SIC computed

for Landsat SIC ranges 0%-5%, 5%-15%, 15%-25%, … , 85%-95%, 95%-100%, red bars one standard deviation of these

mean values, and the red dashed line is the respective linear regression line. Red squares denote the median PMW SIC of the

same Landsat SIC ranges, and the red solid line is the respective linear regression line. The overall mean and median difference

PMW SIC minus Landsat SIC, its standard deviation, and the equation of the linear regression through the individual data

pairs is given at the top, the number N of data pairs and the squared linear correlation coefficient at the bottom of each panel.

9

Figure S02. Same as Fig. S01 but for the Northern Hemisphere.

10

Figure S03. Landsat surface class map of January 29 2014 in the Ross Sea, Southern Ocean, together with Landsat SIC (LSIC)

at 12.5 km, and PMW SIC and their differences to LSIC for the PMW products at 12.5 km grid resolution (see Fig. 9 in the

main manuscript). The surface class map is rotated and enlarged compared to Fig. 9. Colored boxes denote different regions

of interest. Magenta, yellow and orange boxes highlight regions where LSIC is particularly high, exceeding the respective

PMW SIC considerably. Brown boxes highlight regions where LSIC is particularly low, still exceeding SICCI-12km SIC but

being considerably lower than ASI-SSMI SIC. Note the comparably sharp transition from low to high LSIC from the brown

to the yellow box, a feature not shared by the two PMW SIC products shown.

11

Figure S04. Landsat-8 SIC, PMW SIC, and difference PMW SIC minus Landsat SIC (LSIC) for all ten products for a freeze-

up scene in the Fram Strait on April 19, 2015. The Landsat surface class map (top left) shows white: thick/snow covered ice;

grey: thin/new bare ice; black: open water). White and grey pixels are used to compute maps of LSIC at 12.5 km, 25 km and

50 km, respectively (blue: outside Landsat image). A subset of SICCI-12km SIC grid cells shown (top right) illustrates the

array used for the collocation. Panels in the remaining four rows below show PMW SIC and PMW SIC minus LSIC. Land is

flagged brown in the SIC panels and black in the SIC difference panels; it differs between the PMW products. Land masks in

the two bigger maps are from the plotting routine and differ from the land masks of the PMW SIC products. LSIC uses land

masks of the SICCI-2 products.

12

Figure S05. Landsat-8 SIC, PMW SIC, and difference PMW SIC minus Landsat SIC for all ten products for an ice edge scene

in the Bellingshausen Sea, Southern Ocean, on December 24, 2014. See Fig. S04 for a description of the maps shown. Note

that this scene also falls into the set of scenes selected for melt conditions because it is from December.

13

Figure S06. Landsat-8 SIC, PMW SIC, and difference PMW SIC minus Landsat SIC for all ten products for a heterogeneous

ice conditions scene (openings in multiyear pack ice) in the Beaufort Sea, Arctic Ocean, on May 19, 2015. See Fig. S04 for a

description of the maps shown. Note that this scene also falls into the set of scenes selected for melt conditions because it is

from the second half of May.

14

Figure S07. Landsat-8 SIC, PMW SIC, and difference PMW SIC minus Landsat SIC for all ten products for a heterogeneous

ice conditions scene in the Ross Sea, Southern Ocean, on November 23, 2015. See Fig. S04 for a description of the maps

shown. Note that this scene also falls into the set of scenes selected for melt conditions because it is from the second half of

November.

15

Figure S08. Landsat-8 SIC, PMW SIC, and difference PMW SIC minus Landsat SIC for all ten products for a lead / openings

scene north of Franz-Joseph Land, Arctic Ocean, on April 23, 2015. See Fig. S04 for a description of the maps shown.

16

Figure S09. Landsat-8 SIC, PMW SIC, and difference PMW SIC minus Landsat SIC for all ten products for a lead / openings

scene in the Ross Sea, Southern Ocean, on December 8, 2013. See Fig. S04 for a description of the maps shown. Note that this

scene also falls into the set of scenes selected for melt conditions because it is from December.

17

Supplementary material to Section 5

Table S05. Comparison of statistical parameters as listed in Table 8 in the main manuscript or Table S04 but for different sets

of filters (see Subsection 5.3) for the Northern Hemisphere for SICCI-2 and OSI-450 products and Landsat-8. Shown are

values for settings “fully truncated”: the main SIC product with all filters applied, “GT100 off”: only the truncation at near-

100% SIC is switched off (this corresponds “non-truncated” as used throughout the paper), “GT100 / OWF off”: near-100%

SIC and OWF are switched off, “GT100 / LSO off”: near-100% SIC and LSO switched off, and “Fully non-truncated”: no

filters applied, all SIC estimates as retrieved by the algorithms are used.

LS8, NH 2013-15 Diff DiffSDEV Slope Intercept R²

SICCI-12km

Fully truncated -6.2 11.0 0.868 6.1 0.72

GT100 off -4.9 12.1 0.891 5.2 0.68

GT100 / OWF off -4.8 12.2 0.872 7.1 0.68

GT100 / LSO off -4.7 11.9 0.872 7.1 0.69

Fully non-truncated -4.6 11.9 0.853 9.0 0.68

SICCI-25

Fully truncated -4.7 8.2 0.974 -2.4 0.84

GT100 off -4.4 8.5 0.982 -2.7 0.83

GT100 / OWF off -4.3 8.5 0.962 -0.7 0.82

GT100 / LSO off -4.0 7.8 0.934 2.1 0.84


SICCI-50

Fully truncated -3.6 9.0 0.997 -3.3 0.79

GT100 off -3.4 9.1 1.000 -3.5 0.79

GT100 / OWF off -3.3 9.0 0.968 -0.4 0.78

GT100 / LSO off -2.8 7.1 0.943 2.5 0.85


OSI-450

Fully truncated -4.3 9.8 0.779 16.2 0.73

GT100 off -3.9 9.9 0.786 15.9 0.73

GT100 / OWF off -3.9 10.0 0.775 16.9 0.72

GT100 / LSO off -3.9 10.0 0.782 16.3 0.72


18

Table S06. As Table S05 but for the Southern Hemisphere.

LS8, SH 2013-15 Diff DiffSDEV Slope Intercept R²

SICCI-12km

Fully truncated -5.3 13.3 0.915 2.1 0.78

GT100 off -4.3 13.8 0.931 1.6 0.77

GT100 / OWF off -4.1 13.7 0.912 3.4 0.77

GT100 / LSO off -4.1 13.3 0.921 2.7 0.78


SICCI-25

Fully truncated -5.9 13.7 0.969 -3.3 0.77

GT100 off -5.6 13.7 0.976 -3.5 0.77

GT100 / OWF off -5.4 13.6 0.959 -1.9 0.76

GT100 / LSO off -4.7 11.9 0.952 -0.6 0.81


SICCI-50

Fully truncated -6.8 16.0 1.033 -9.6 0.72

GT100 off -6.5 16.2 1.040 -9.9 0.71

GT100 / OWF off -6.0 16.0 0.988 -5.0 0.70

GT100 / LSO off -3.9 11.5 0.953 0.1 0.81


OSI-450

Fully truncated -5.3 13.5 0.827 9.5 0.73

GT100 off -5.1 13.7 0.832 9.3 0.73

GT100 / OWF off -5.0 13.7 0.816 10.8 0.72

GT100 / LSO off -4.9 13.5 0.818 10.7 0.73


supplementary material table s01

Documents